Re: [PATCH v4] memory: Optimize replay of guest mapping

[Peter Maydell]

On Thu, 20 Apr 2023 at 04:13, Duan, Zhenzhong <zhenzhong.duan@intel.com> wrote:
> >From: Peter Maydell <peter.maydell@linaro.org>

> >Question: is it OK for an implementation of this method to call the notifier 
> >for
> >translations that are in the IOMMU and which are not in the scope of the
> >notifier (ie which are outside the intersection) ? Or must it specifically 
> >restrict
> >itself to only calling the notifier for translations which are inside the 
> >notifier's
> >range ?
> In the calling path to notifier->notify(), memory_region_notify_iommu_one()
> ensures passing an IOMMUTLBEntry in the notifier's range.
> I think it's ok for an implementation of replay() to walk over the entire
> IOMMU MemoryRegion because unrelated entries are filtered as above.
> It's just less optimistic which this patch try to address for intel_iommu.

OK, great.

-- PMM